[Paper] Adaptive Resource Orchestration for Distributed Quantum Computing Systems

Source: arXiv - 2512.24902v1

Overview

The paper introduces Modular Entanglement Hub (ModEn‑Hub), a new architecture for linking many quantum processing units (QPUs) into a distributed quantum‑HPC system. By centralizing entanglement generation and adding a real‑time orchestration layer, ModEn‑Hub dramatically improves the success rate of non‑local quantum operations, making large‑scale quantum computing more practical on near‑term hardware.

Key Contributions

• Hub‑and‑spoke photonic interconnect that provides on‑demand high‑fidelity Bell pairs to heterogeneous QPUs.
• Quantum network orchestrator that schedules teleportation‑based gates, launches parallel entanglement attempts, and maintains a small “ebit cache” for opportunistic reuse.
• Monte‑Carlo evaluation framework that simulates realistic photon loss, tight timing budgets, and a range of system sizes (1–128 QPUs).
• Demonstrated performance boost: ~90 % teleportation success vs. ~30 % for a naive sequential approach, with a modest increase in average entanglement attempts (≈10–12 vs. ≈3).
• Open‑source, reproducible code for the simulation, enabling other researchers and engineers to extend the study.

Methodology

1. Architecture Modeling – The authors model a hub that houses entanglement sources and a shared quantum memory (the “ebit cache”). Each QPU connects to the hub via a photonic link that suffers realistic loss.
2. Orchestration Policy – The control plane implements a logarithmically scaled parallelism strategy: when a non‑local gate is requested, the orchestrator fires multiple entanglement attempts in parallel, doubling the number of attempts each round until a success is cached.
3. Monte‑Carlo Simulation – For each configuration (number of QPUs, loss parameters, round‑time budget), 2,500 independent trials are run. The simulation tracks:
   • Number of entanglement attempts per gate,
   • Success probability of teleportation,
   • Cache hit rate.
4. Baseline Comparison – A naive sequential baseline attempts a single entanglement generation per gate, waiting for success before moving on.

All components are implemented in a lightweight Python package, with parameters exposed for easy tweaking.

Results & Findings

• Success Rate: The orchestrated policy keeps teleportation success above 90 % even as the number of QPUs grows, whereas the baseline collapses toward 30 % due to cumulative loss and timing constraints.
• Resource Overhead: The higher success comes with roughly 3–4× more entanglement attempts per gate, but these attempts are cheap (photon generation) and can be parallelized across the hub’s hardware.
• Cache Effectiveness: The ebit cache yields a hit rate of ~45 % after the first few rounds, further reducing latency for subsequent gates.

Overall, the study provides clear evidence that adaptive orchestration outweighs the modest extra entanglement cost, enabling scalable distributed quantum computation.

Practical Implications

• Quantum Cloud Providers: A hub‑centric design can be rolled out as a service layer that abstracts away the messy details of entanglement generation, letting users submit quantum circuits without worrying about network reliability.
• Compiler & Runtime Integration: Existing quantum compilers can target the orchestrator’s API to emit teleportation‑based non‑local gates, while runtime systems can exploit the ebit cache for latency‑critical subroutines (e.g., error‑correction cycles).
• Hardware Planning: Engineers can allocate modest additional photonic sources at a central node rather than provisioning each QPU with its own high‑rate source, reducing cost and simplifying calibration.
• Hybrid Classical‑Quantum Workloads: The orchestrator’s scheduling logic can be extended to co‑schedule classical control messages, making end‑to‑end pipelines (data preprocessing → distributed quantum kernel → post‑processing) more deterministic.

In short, ModEn‑Hub offers a software‑defined networking approach for quantum hardware, analogous to SDN in classical data centers, and could become a foundational building block for future quantum‑HPC platforms.

Limitations & Future Work

• Simplified Loss Model: The simulation assumes static loss rates per link; real‑world fiber fluctuations and component aging could affect performance.
• Fixed Round Budget: The study uses a tight, uniform timing budget per gate; adaptive timing (e.g., longer windows for high‑loss links) is not explored.
• Scalability of the Hub: While the hub is assumed to have enough photonic sources to support parallel attempts, the physical limits (laser power, detector saturation) need quantitative analysis.
• Security & Fault Tolerance: The paper does not address how the orchestrator handles malicious or faulty QPUs, nor does it explore redundancy mechanisms for the hub itself.

Future research directions include integrating dynamic loss estimation, extending the orchestrator to multi‑hub topologies, and prototyping the architecture on emerging photonic‑integrated platforms.

Authors

• Kuan-Cheng Chen
• Felix Burt
• Nitish K. Panigrahy
• Kin K. Leung

Paper Information

• arXiv ID: 2512.24902v1
• Categories: quant-ph, cs.DC
• Published: December 31, 2025
• PDF: Download PDF